From owner-freebsd-current  Mon Oct 19 23:18:14 1998
Return-Path: <owner-freebsd-current@FreeBSD.ORG>
Received: (from majordom@localhost)
          by hub.freebsd.org (8.8.8/8.8.8) id XAA12415
          for freebsd-current-outgoing; Mon, 19 Oct 1998 23:18:14 -0700 (PDT)
          (envelope-from owner-freebsd-current@FreeBSD.ORG)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.15.68.22])
          by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id XAA12405
          for <current@FreeBSD.ORG>; Mon, 19 Oct 1998 23:18:09 -0700 (PDT)
          (envelope-from bde@godzilla.zeta.org.au)
Received: (from bde@localhost)
	by godzilla.zeta.org.au (8.8.7/8.8.7) id QAA00889;
	Tue, 20 Oct 1998 16:17:42 +1000
Date: Tue, 20 Oct 1998 16:17:42 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199810200617.QAA00889@godzilla.zeta.org.au>
To: gibbs@narnia.plutotech.com, mike@smith.net.au
Subject: Re: vote:? reversion of BDE change please?
Cc: current@FreeBSD.ORG
Sender: owner-freebsd-current@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

>> The change is actually in the right direction, as the goal is to remove 
>> the prototypes from global scope completely.  Instead, ISA device 
>> drivers should register their interrupt handlers at attach time.  I 
>> don't know if there's an example of this at the present time.
>
>The CAM ISA SCSI drivers set their own vectors and thus do not need
>entries in isa_devices.h.

Examples for doing it not-quite-right are trivial.  For old drivers, it
it essentially takes a 1 line assignment:

diff -c2 syscons.c~ syscons.c
*** syscons.c~	Fri Oct  2 14:44:31 1998
--- syscons.c	Tue Oct 20 15:53:49 1998
***************
*** 240,243 ****
--- 240,244 ----
  /* prototypes */
  static int scattach(struct isa_device *dev);
+ static ointhand2_t scintr;
  static int scparam(struct tty *tp, struct termios *t);
  static int scprobe(struct isa_device *dev);
***************
*** 646,649 ****
--- 647,651 ----
  #endif
  
+     dev->id_intr = (inthand2_t *)scintr;
      scinit();
      sc_flags = dev->id_flags;
***************
*** 852,856 ****
  }
  
! void
  scintr(int unit)
  {
--- 854,858 ----
  }
  
! static void
  scintr(int unit)
  {

Also, the declaration of scintr() must be changed to `#define scintr NULL'
in isa_device.h so that old config files don't break.  For new drivers,
just don't put interrupt vectors in config files or declarations for
interrupt handlers in isa_device.h.

I have converted about 50 drivers as above.

New drivers should also use `inthand2_t' interrupt handlers and not
cast to (inthand2_t *).  There is not enough infrastructure for doing
this cleanly - it currently takes larger, uglier code than assigning
to id_intr.

I have converted about 5 drivers to use a NetBSDish interrupt attach
function.  The larger, uglier code goes away, something like this:

diff -c2 bt_isa.c~ bt_isa.c
*** bt_isa.c~	Tue Oct 13 19:42:04 1998
--- bt_isa.c	Tue Oct 13 19:42:20 1998
***************
*** 43,47 ****
  static	int bt_isa_probe __P((struct isa_device *dev));
  static	int bt_isa_attach __P((struct isa_device *dev));
- static	void bt_isa_intr __P((void *unit));
  
  static	bus_dma_filter_t btvlbouncefilter;
--- 43,46 ----
***************
*** 196,199 ****
--- 208,212 ----
  	void		 *filter_arg;
  	bus_addr_t	 lowaddr;
+ 	int	result;
  
  	bt = bt_softcs[dev->id_unit];
***************
*** 282,298 ****
  	}
  
! 	return (bt_attach(bt));
! }
! 
! /*
!  * Handle an ISA interrupt.
!  * XXX should go away as soon as ISA interrupt handlers
!  * take a (void *) arg.
!  */
! static void
! bt_isa_intr(void *unit)
! {
! 	struct bt_softc* arg = bt_softcs[(int)unit];
! 	bt_intr((void *)arg);
  }
  
--- 295,301 ----
  	}
  
! 	result = bt_attach(bt);
! 	isa_intr_establish(dev, &bio_imask, bt_intr, bt);
! 	return (result);
  }
  
However, I think this doesn't actually work for drivers that that hack
on dev->id_intr (pnp and maybe pccard drivers).

Bruce

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message